Ahmed, Rose, Figliozzi & Jacob
1
Commuter Cyclist’s Sensitivity to Changes in Weather: Insight from Two Cities with
Different Climatic Conditions
Farhana Ahmed (Corresponding author)
Institute of Transport Studies, Department of Civil Engineering
Monash University
Victoria, Australia 3800
Phone: +61 3 9905 1848
Fax : +61 3 9905 4944
E-mail: farhana.ahmed@monash.edu
Geoffrey Rose
Institute of Transport Studies and Monash Sustainability Institute
Monash University
Victoria, Australia 3800
Phone: +61 3 9905 4959
Fax : +61 3 9905 4944
E-mail: geoff.rose@monash.edu
Miguel Figliozzi
Civil and Environmental Engineering
Portland State University
Portland, OR, 97207-0751
Email: figliozzi@pdx.edu
Christian Jakob
Atmospheric Science, School of Mathematical Sciences
Monash University
Phone: +61 3 99054461
Email: Christian.Jakob@monash.edu
Word Count: 5710 words + 4 Figures (4*250 word equivalents) + 7 Tables (7*250 word
equivalents) = 8460 equivalent words.
Revised paper submitted on 15November, 2011
TRB 2012 Annual Meeting
Paper revised from original submittal.
Ahmed, Rose, Figliozzi & Jacob
2
ABSTRACT
This study examines the relationship between various weather conditions and commuter
bicyclist volume in two cities (Portland, USA and Brisbane, Australia), which fall into
different climatic zones. Investigating the variation in day-to-day bicycle ridership can
help to understand factors influencing demand and in particular how base climatic
conditions may condition bicyclist’s responsiveness to changes in weather and climate.
Temporal variations in bicycle usage and key weather parameters (temperature and
rainfall) are analyzed. Ridership counts and weather data are then used to develop an
aggregate demand model that provides quantitative insight into the effects of weather on
bicyclist volumes. The results indicate that daily bicyclist volume varies across hours of
the day as well as days of the week. Both temperature and rainfall are found to have a
significant influence on daily bicyclist volume but with different degrees of sensitivity in
these two cities which correspond to their base climates. The results are discussed in view
of their implications for government strategies that seek to increase the role of bicycle in
urban areas.
TRB 2012 Annual Meeting
Paper revised from original submittal.
Ahmed, Rose, Figliozzi & Jacob
3
INTRODUCTION
Policy makers are promoting bicycling as a mode of transportation since it reduces traffic
congestion, energy consumption and greenhouse gas emissions and enhances health
outcomes (1, 2, 3) and therefore offers sustainable economic, environmental and social
benefits (4). At the same time, changes in weather are high on the agenda in cities around
the world today because the intensity and frequency of extreme weather conditions are
expected to increase as a result of climate changes (5). Since cyclists are directly exposed
to the weather this paper seeks to quantify the impact of changes in weather on bicyclist’s
travel behavior.
There are many factors that affect demand for bicycling although research
consistently highlights the importance of adequate infrastructure (6). Apart from investing
in infrastructure, such as off-road paths and bicycle lanes to accommodate the growing
number of cyclists and to facilitate even more growth, governments have regularly
sponsored events to encourage bicycle commuting (e.g. ‘Ride to Work Day’ (7)). Seasonal
variations in ridership (8, 9) suggest that there are many fair weather utilitarian cyclists and
this presents a challenge from the perspective of gaining the maximum benefit from the
investments in bicycle infrastructure targeted on promoting this mode of transport. Policies
could be undermined by the influence of weather; particularly where the target is to
increase women’s cycling (2), if women are more likely to be deterred from riding in
adverse weather.
Following global policy, Portland, Oregon and Brisbane, Queensland has been
encouraging bicycling. Portland is well known for being one of the most bicycle friendly
cities in the USA. In 2008 Portland became the first major city in the USA designated as a
Platinum-level Bicycle Friendly Community by the League of American Bicyclists (10,
11). Likewise, in Queensland, cycling has been encouraged by allocating funds to develop
cycle networks, through supporting various cycle events (12) and developing world-class
end-of-trip facilities to promote commuter cycling (13). While bicycle count data has been
complied for more than ten years (14), little research has explored the temporal variability
in bicycle volume and in particular, how commuter cyclists of Portland and Brisbane are
influenced by weather conditions.
Portland and Brisbane fall into different climatic zones; Portland’s climate is more
temperate mediterranean climate whereas Brisbane experiences a tropical climate (15). As
noted earlier, weather and climate has the potential to influence cycling. Thus, this paper
seeks to quantify the influence of day to day changes in weather on bicyclist volume in
these two cities whose base climatic condition are different. This analysis can also provide
insight into weather corrections which could be useful in adjusting counts made at one
point in time (e.g. census journey to work data) to account for regional differences in
weather/climate.
The structure of this paper is as follows. Section two provides a review of the
relevant literature. The review focuses on the effect of changing weather conditions on
travel behavior and identifies the sources of data and the methodologies which have been
used to develop understanding of those relationships. The following section examines
temporal variation of bicyclist volume in Portland and Brisbane. It also explores how
various weather parameters specifically, temperature and rainfall, vary over the year. Data
from Portland and Brisbane are then used to calibrate aggregate bicycle demand models
that incorporate a range of explanatory variables including weather. The final section
presents the conclusions of the research and identifies future research needs and directions.
TRB 2012 Annual Meeting
Paper revised from original submittal.
Ahmed, Rose, Figliozzi & Jacob
4
INSIGHT FROM THE LITERATURE
There is limited research which explores how travel behavior is influenced by weather.
Nankervis (16) examined how weather and climate affects bicycle commuters in
Melbourne, Australia. He identified a decrease in cycling over the winter months. Heavy
rain was the biggest deterrent for the cyclists to ride with 67% of the respondents
indicating they would be deterred from riding in heavy rain. Among these respondents who
did not ride (67%), almost all of them (90%) indicated that they still made the trip but used
an alternative mode. Thomas et al (17) identified that temperature caused greatest variation
and wind caused the least variation in bicycling demand in Netherlands. More recently,
Lewin (18) confirmed that the impact of temperature on daily bicyclist volume in Boulder,
CO was non linear with the optimum riding temperature estimated to be 32.20C. Moreno
and Nosal (1) investigated how bicycle usage in Montreal, Canada is impacted by various
weather conditions. Their analyses found that precipitation, temperature and humidity
influence bicycle ridership. When other factors are controlled, a 100% increase in
temperature increases the ridership by 43-50%. However, temperature greater than 280C
and humidity greater than 60%, reduced the ridership which confirms a non- linear effect.
Precipitation was also found to have both a direct and a lagged effect on ridership. As a
result, bicyclist volume in a particular hour is not only affected by the presence of rain in
that hour but also affected if there was rain in previous hours. Other research (19)
examined ridership sensitivity to weather in Portland, Oregon, USA. Analyses of six
months of data from Portland, Oregon indicated that a 10C rise in temperature increases the
volume of daily bicyclists by between 3% to 6% whereas each1mm increase of
precipitation decreases the volume by around 4%.
Both Richardson (20) and Phung and Rose (21) explored how weather variations
affect bicycle ridership in Melbourne, Australia. Rain was identified as the most influential
weather parameter which significantly decreased commuting cyclist volumes. Both of
these studies found that rainfall has a non-linear effect. Richardson (20) identified that
daily rainfall of around 8 mm, reduces cyclist volumes by about 50% compared to days
when there is no rain. In contrast, Phung and Rose (21) found that light rain (defined as
daily rainfall less than 10 mm) deterred between 8 and 19% of all cyclists while heavy rain
(defined as daily rainfall greater than 10 mm) deterred about one-third more (13 to 25%).
Air temperature has been identified to have a non-linear and non-symmetrical relationship
on commuter cyclist volume with the volume of riders decreasing immediately after the
ideal riding temperature (20, 21). Phung and Rose (21) identified the ideal riding air
temperature to be about 280C whereas Richardson’s (20) analysis identified the optimal air
temperature for riding to be 250C. Wind effects were detected for most of the sites in
Melbourne studied by Phung and Rose (21), but ridership on the Bay Trail, which runs
along the exposed coast of Port Phillip Bay, was the most sensitive to wind change.
Table 1 summarizes the general nature of the conclusions reached from studies
focused on different geographic locations which have a range of background climatic
conditions. The literature suggests that across the locations which have been examined,
cyclists have been found to be sensitive to weather although that sensitivity varies across
different weather parameters. Temperature and precipitation are the most important
influences on bicyclist volume. Wind was found to be deterrent in some places but mostly
identified as being less influential.
TRB 2012 Annual Meeting
Paper revised from original submittal.
Ahmed, Rose, Figliozzi & Jacob
5
TABLE 1 Different Location’s Sensitivity to Weather Parameters
Location
Weather parameter
Climate
Reference
Temperature Precipitation Wind
Melbourne,



Temperate
Nankervis
Australia
Highly
Least
oceanic
(16)
influential
influential climate
Netherlands


Temperate
Thomas et
Highly

oceanic
al (17)
influential
climate
Boulder, CO


Warm oceanic Lewin (18)
Highly
climate
influential
City of



Warm
Moreno
Montreal,
Continental
and Nosal
Canada
climate
(1)
Portland,


Temperate
Rose et al
Oregon
Highly
Mediterranean (19)
influential
climate
Melbourne,


Temperate
Richardson
Australia
Highly
oceanic
(20)
influential
climate
Melbourne,



Temperate
Phung and
Australia
Highly
oceanic
Rose (21)
influential
climate
Melbourne,



Temperate
Ahmed et
Australia
oceanic
al (9)
climate
Researchers have drawn on a range of data types and sources to explore how travel
behavior is related with weather. Table 2 summarizes some of the widely used data
sources. It is found that travel behavior data is exclusively collected by questionnaire
surveys but in the case of ridership data, it has been collected both automatically and
through observational surveys.
TRB 2012 Annual Meeting
Paper revised from original submittal.
Ahmed, Rose, Figliozzi & Jacob
Data
Travel
behavior
data
Ridership
data
6
TABLE 2 Different Data Types and Sources
Data Sources
References
Questionnaire
Survey
Automatic Counting
System
Khattak and Palma (22)
Pneumatic tube
Detector loops
Weather
data
Observational Survey
Usually the weather data is provided by
the relevant local meteorological
organizations
Thomas et al (17), Rose et al
(19)
Moreno and Nosal (1), Ahmed et
al (9), Lewin (18), Rose et al
(19), Phung and Rose (21),
Nankervis (16)
Ahmed et al (9),
Nankervis (16), Thomas et al
(17), Lewin (18), Rose et al (19),
Phung and Rose (21),
In recent years, most of the research has relied on automatically counted hourly
ridership data. The advantage of an automatically counting approach is that it gives the
data for a continuous period whereas manually collected data may represent a specific
period of a year, or at the very least a much more limited time period. The local Bureau of
Meteorology is usually a rich source of weather data with information available in either
an hourly or 24 hour aggregate format. Depending on the requirement of the research, both
hourly and daily data have been analyzed in different studies.
When seeking to understand weather or seasonal patterns in ridership data, a first
level of analysis often involves use of plots or descriptive statistics to understand patterns
in the data (16, 20, 23). Higher level analysis usually focuses on development of a
statistical model to explore the relationship between a range of explanatory variables and a
dependent variable (usually ridership volume). Models that have been employed include
time series models (24), linear and non-linear regression models (17, 21) and ordered
probit models (22).
The research described above is expanding understanding of how travel behavior is
influenced by changing weather and climate. This paper adds to that knowledge base by
combining weather effects in two cities with differing base climatic scenarios.
TEMPORAL VARIATION IN RIDERSHIP AND WEATHER
We now turn attention to the two cities considered in the analysis: Portland, Oregon (USA)
and Brisbane, Queensland (Australia). The data available from those two locations is
described, and then the day to day variability in key parameters is quantified.
Data
Portland's bridges act as feeders to carry commuters and students from the neighborhoods
of the east into the downtown area and beyond. Bicycle volumes have been monitored on
these bridges since the early 1990’s through automatic hose counts. However, the data
utilized in this research was collected using state of the art bicycle sensors of high accuracy
(25) which were installed in August 2009. Among the bridges that carry pedestrian and
cyclists to the downtown area, the Hawthorne Bridge has the greatest use and carries more
TRB 2012 Annual Meeting
Paper revised from original submittal.
Ahmed, Rose, Figliozzi & Jacob
7
bicycle traffic than all other Portland bridges combined. Because of its importance in the
network, and the availability of quality data, the Hawthorne Bridge is considered in the
analysis reported here. Previous research (19) which examined the impact of weather on
Portland cyclists could only draw on data covering a six month period. As a result of
additional data becoming available, the analysis reported here considers a total period in
excess of twelve months.
Automatic bicycle counts have been conducted in Queensland for more than ten
years (14) and they provide continuous, directional data at 15 min to 1 hour intervals
(varies from site to site). The site considered here for analysis is located on a dedicated
bicycle path as it passes through Mowbray Park, Brisbane (an inner suburb) which was
found to be predominantly a commuter trail in earlier research (26). Moreover, it is a
permanent count location and had less missing data compared to other candidate locations
in Brisbane.
Overall, the data used here to analyze the variation in ridership covered the period
September 2009 to December 2010 and thus, the seasonal variation for both sites is
captured. The corresponding weather data for the two sites was also obtained from the
relevant meteorological organizations.
Variation in Ridership
The Average Annual Daily Traffic (AADT) values, computed separately for weekdays and
weekends and public holidays, for the two sites are summarized in Table 3. The table
indicates the variability in usage. Ratios of AADT of weekday to weekend and public
holiday are also shown to identify the predominant functionality of each site.
TABLE 3 Average Annual Daily Traffic (AADT) By Weekday/Weekend and
Public Holiday
Site name
AADT
Ratio*
Predominant type of
trail**
Week Weekend and
day
public holiday
Hawthorne Bridge
4478
2023
2.21
Commuter
Mowbray Park
488
165
2.95
Commuter
*Ratio = Weekday AADT/ Weekend and public holidays AADT
**Type of trail: Commuter trail, ratio >1
As shown in the above table, both sites have higher bicyclist volume on weekdays than
weekends and public holidays. For both sites the ratio of weekday/weekend is above 1.
Considering the magnitude of the ratios it can be concluded that both sites are
predominantly catering for commuter cyclists (21).
Figure 1 shows the average weekday hourly bicycle volume for each site. Bicyclist
volume varies by time of day with similar patterns across the sites. On the basis of AADT,
Hawthorne Bridge serves an order of magnitude more commuter cyclists than the
Mowbray Park location however peak hour counts only differ by a factor of three to four.
As highlighted in Figure 1, here is clearly more inter-peak bicycle traffic on the Hawthorne
Bridge than at the Mowbray Park site.
TRB 2012 Annual Meeting
Paper revised from original submittal.
Ahmed, Rose, Figliozzi & Jacob
8
FIGURE 1 Average hourly bicyclist volume across sites
At both sites, ridership predictably peaks twice per day, in morning (6:00 to 9:00)
and in evening (16:00 to 19:00). The identified peak periods again confirm that the sites
are primarily used by commuters.
Variation in Key Weather Parameters
Monthly variation in the key weather variables, temperature and rainfall (18), is
illustrated in Figures 2 and 3 respectively. The values in the figures reflect the average
over two years (2009 and 2010) for Portland and three years (2008, 2009 and 2010) for
Brisbane. A clear contrast in temperature range between Portland and Brisbane can be seen
in Figure 2. During summer in Portland, temperature ranges from around 150C to 270C
whereas it is 230C to 280C in Brisbane. The variation is more pronounced during winter.
Temperature ranges from 120C to 220C in winter in Brisbane whereas Portland experiences
lower temperatures and a wider temperature range of -40C to 110C during the same season.
The plots of average monthly temperature highlight that temperature peaks in August in
Portland, whereas in Brisbane, reflecting its southern hemisphere location, experiences its
lowest temperatures at that time of year.
FIGURE 2 Monthly variation of average daily temperature
TRB 2012 Annual Meeting
Paper revised from original submittal.
Ahmed, Rose, Figliozzi & Jacob
9
Figure 3 depicts the variation in rainfall over the year where daily rainfall averaged
on a monthly basis is plotted. The upper plots show actual level of rainfall in Portland and
Brisbane whereas the lower plots indicate indexed values. To calculate the indexed value,
the average rainfall is normalized with respect to the December average since the highest
amount of rainfall in both cities falls in December. The value in each month therefore
reflects the average daily rainfall in that month relative to the December base. For
example, July rainfall in Portland is about 10 % of that which falls in December.
FIGURE 3 Monthly variation of total daily rainfall
Maximum rainfall of around 5 mm is observed during winter months in Portland.
In Brisbane, rainfall peaks during summer months when the maximum average daily
rainfall is about 6mm. In both cities, the driest months are July-August when Portland gets
only about 5% of the maximum rainfall in December whereas for Brisbane it is around
25%.
The variation in temperature and rainfall are measured by the Coefficient of
Variation (COV) and summarized in Table 4 for both locations. The COV for temperature
is much higher in Portland but the two cities experience a similar degree of variation in
rainfall.
TABLE 4 Degree of Variation in Temperature and Rainfall
Location
Coefficient of Variation
Temperature
TRB 2012 Annual Meeting
Rainfall
Portland
.43
.50
Brisbane
.18
.52
Paper revised from original submittal.
Ahmed, Rose, Figliozzi & Jacob
10
MODELING COMMUTER CYCLIST BEHAVIOR
To estimate the effect of weather and other potential factors into bicyclist volume a
regression modeling approach is adopted in this study as this approach has previously
provided valuable insight (1, 9, 18, 19, 21). As the analysis reported here focuses on
commuter cyclist behavior, Saturdays and Sundays are excluded from the analysis. Public
holidays are included in order to estimate how much public holidays influence the
weekday ridership.
To examine the variation in bicycle usage and the factors contributing to those
variations, a log linear model is employed. Different combinations of explanatory variables
have been used in the model to identify which combination yield the best fit to the
ridership data. Equation 1 summarizes the formulation of the underlying model.
Model
(1)
Where;
Q = Daily total (24 hour) bicycle volume
i = Site index
t = Time index
DOW = DOW variable is coded using four dummy variables (Monday through
Thursday) with the base case corresponding to Friday (when the Day of Week dummies
for Monday through Thursday are equal to zero).
TIME = a variable which increments to reflect the day when the data was
collected (incrementing from the start to the end of the series of data). This variable is
used to capture any time based growth effect.
PUB = Public holiday, coded as a dummy variable (1 for a public holiday and 0
otherwise)
TEMP = Average daily (24 hour) temperature (Celsius).
RAIN = Daily total (24 hour) rainfall (mm)
Three alternative functional forms are used for TEMP and RAIN variable:
1. Continuous variable – linear effect
2. Continuous variable – non-linear effect
TRB 2012 Annual Meeting
Paper revised from original submittal.
Ahmed, Rose, Figliozzi & Jacob
11
3. Categorical representation
The third functional form (categorical representation) is considered to reflect
differences in the underlying climate of the cities. In this representation,
TEMP variable is coded using three dummy variables corresponding to a ‘Cool
Day’, a ‘Warm Day’ and a ‘Mild Day’ with the base case (all temperature dummies set
to zero) corresponding to a ‘Very Warm Day’. The RAIN variable is coded using two
dummy variables for ‘Light rain’ and ‘Heavy rain’ with the base case corresponding to
‘No rain’.
The categorization of both ‘TEMP’ and ‘RAIN’, shown in Table 5, is based on
analysis of quartiles from the respective distributions. Each quartile is given a linguistic
interpretation for temperature (i.e. Cool, Mild, Warm or Very Warm Day) while the
fourth quartile (i.e. the highest 25% of the daily rainfall totals) is used for defining
heavy rain days. The categorization of the variables ‘TEMP’ and ‘RAIN’ corresponds to
how each area experiences the variation in temperature and rain and so, therefore
reflects differences in the base climatic conditions of the two cities.
TRB 2012 Annual Meeting
Paper revised from original submittal.
Ahmed, Rose, Figliozzi & Jacob
12
TABLE 5 Categorization of Variables ‘TEMP’ and ‘RAIN’
Weather
parameter
Category
Portland
Brisbane
Temperature
Cool day
< 7.20C
< 18.20C
7.20C to 10.60C
18.20C to 21.70C
10.60C to 16.10C
21.70C to 24.30C
>16.10C
> 24.30C
No rain
0 mm
0 mm
Light rain
Up to 3.8 mm
Up to 1.4 mm
> 3.8 mm
> 1.4 mm
(First quartile)
Mild day
(Second quartile)
Warm day
(Third quartile)
Very warm day
(Fourth quartile)
Rainfall
(up to the third
quartile)
Heavy rain
(Fourth quartile)
As Equations 1 is a log-linear formulation, the coefficients of the continuous
variables are directly interpreted as the percentage change in the dependent variable (daily
bicycle volume) as a function of a change in the explanatory variable. For example, a
coefficient of 0.08 on the ‘TEMP’ variable in functional form 1 would imply that an
additional 10C of temperature would cause an 8% increase in daily bicycle volume. In
contrast, the effect of the dummy variables is measured as effect= eβ– 1, where β = coefficient of the explanatory variable (21). The interpretation is relative to the base case. For
example, coefficient of 0.13 for Tuesday would be interpreted as the volume of a Tuesday
is 14% (e0.13-1) higher than that of Friday (since Friday is the reference day for coding the
data). The analysis focuses on the effects of temperature and rainfall since those weather
variables were found to be the most important in explaining variations in ridership in
earlier studies (16, 18, 21). Correlation among the explanatory variables was examined but
no significant correlations were observed. The highest correlation of -0.37 was observed
between ‘TIME’ and ‘Mild Day’.
TRB 2012 Annual Meeting
Paper revised from original submittal.
Ahmed, Rose, Figliozzi & Jacob
13
MODELLING RESULTS AND DISCUSSIONS
The modeling results for Portland and Brisbane are presented in Table 6 and Table 7
respectively. The estimated coefficients and the Coefficient of Determination (R2), a
measure of ‘Goodness of fit’ for OLS regression, are also given in these tables. The shaded
boxes correspond to significant variables at a 95% confidence level.
Portland Results
For Portland models, the R2 value ranges from 0.67 to 0.74 across the models. Thus,
around 70% of the variation in daily bicyclist volume has been explained by these models.
The poorest fit is for model number 3 where temperature and rain are both included as
categorical variables.
TRB 2012 Annual Meeting
Paper revised from original submittal.
Ahmed, Rose, Figliozzi & Jacob
14
TABLE 6 Regression Model Results for Portland
Model Number
Explanatory Variable
Time based growth
Mon
Tue
Day of week
Wed
Thu
Public holiday
1
2
3
4
-0.00019
(-1.58)
.14
(3.09)
.16
(3.38)
.17
( 3.66)
.13
( 2.78)
-.9
( -12.6)
-0.00018
( -1.6)
.13
(3.09)
.15
(3.36)
.16
(3.80)
.12
(2.74)
-.9
(-13.25)
-0.0001
(-0.74)
0.13
(2.77)
0.12
(2.51)
0.15
(3.02)
0.10
(2.09)
-0.99
(-13.08)
-0.66
(-13.69)
-0.36
(-6.85)
-0.16
(-3.49)
-0.0002
(-1.74)
0.12
(2.83)
0.13
(3.04)
0.15
(3.42)
0.10
(2.21)
-0.94
(-13.85)
.045
(17.76)
.08
(11.87)
0.08
(11.63)
-0.002
(-5.51)
-.02
(-7.73)
-0.002
(-5.44)
-0.04
( -7.50)
0.0008
( 4.11)
Cool day
Mild day
Warm day
TEMP
TEMP2
RAIN
RAIN2
Light rain
Heavy rain
R2
.71
.74
-0.02
(-0.60)
-0.26
(-6.45)
.67
-0.10
(-2.69)
-0.35
(-9.31)
.74
KEY: The associated t stats are presented in parenthesis. The critical values for t stats are for a 95%
confidence interval. Variables which are significant at a 95%confidence level are lightly shaded.
Time based growth is not found to be statistically significant in any model. The
changes in volume across different days of the week show the same trend for all the
models. In each case the reference day is Friday. Bicyclist volume reaches its peak in the
middle of the week with Wednesday recording the highest volume which is around 17% (e
(-.16)
-1) higher than that on Friday. Volume declines as the week progresses with lowest
volume on Friday. On Public holidays, the daily bicyclist volume decreases by around 60%
(e (-.94) -1).
TRB 2012 Annual Meeting
Paper revised from original submittal.
Ahmed, Rose, Figliozzi & Jacob
15
Temperature was found to have significant effect on bicyclist volume in all models.
From model 3 where temperature is specified as a dummy variable, it is estimated that very
warm days (those in the top quartile of temperatures >16.10C) correspond to the highest
bicyclist volume. Warmer weather increases bicyclist volume. However, when temperature
is specified as continuous variable, a non- linear effect is identified in model 2 and 4which
is illustrated in Figure 4.
FIGURE 4 Effect of temperature on bicyclist volume
An optimum riding temperature of 240C is indicated by Figure 4. The result shows
that when the temperature reaches to the optimum level, volume increases by around 400%
from that at -70C (the recorded minimum temperature). However, the effect of temperature
drops off when it is greater than 240C
The effect of heavy rain is found to be statistically significant in models 3 and 4
though the effect of light rain is found to be statistically significant only in model 4. The
greatest impact comes from heavy rain which decreases volume by 29% which is about
three times higher than the effect of light rain. When the effect of rain is treated as
continuous variable, it is observed that each 1 mm increase of rain reduces the bicyclist
volume by around 3%.
Brisbane Results
The Brisbane model results are presented in Table 7. The first four models have the same
specifications as for Portland. Model 5 considers a different rainfall effect. The availability
of hourly rainfall data for Brisbane made it possible to examine the impact of early
morning rain on daily ridership. The dummy variable ‘Presence of rain_5am-8am’ in
model number 5 indicates days when there was rain during 5am to 8am period.
The R2 value ranges from 0.56 to 0.61 across the models. Thus, around 60% of the
variation in daily bicyclist volume has been explained by these models. The discussion
focuses first on models1 to 4 and then returns to consider model 5.
The negative and statistically significant (in all except model 4) time based growth
coefficients highlight that ridership at the Mowbray Park site is declining over time. Over
the analysis period, the site has experienced a reduction of about 10 % (-.0003*365) per
annum in daily bicycle volume.
TRB 2012 Annual Meeting
Paper revised from original submittal.
Ahmed, Rose, Figliozzi & Jacob
16
TABLE 7 Regression Model Results for Brisbane
Model Number
1
2
-0.0003
-0.0003
(-2.4)
(-2.16)
3
-0.0003
(-2.09)
4
-0.0002
(-1.86)
5
-0.0003
(-2.86)
Mon
.24
(4.55)
.25
(5.14)
0.23
(4.49)
0.24
(4.60)
0.29
(5.85)
Tue
.33
(6.42)
.34
(6.87)
0.34
(6.61)
0.34
(6.65)
0.33
(6.72)
Wed
.30
(5.8)
.30
(6.15)
0.30
(5.87)
0.31
(5.93)
0.31
(6.16)
Thu
.25
(4.88)
.25
(5.10)
0.22
(4.32)
0.22
(4.31)
0.26
(5.30)
-1.25
(-13.91)
-1.22
(-14.29)
-1.28
(-14.49)
-1.29
(-14.50)
-1.23
(-14.33)
.08
(1.51)
0.07
(1.37)
-0.01
(-1.63)
-0.002
(-1.57)
-0.002
(-1.37)
Explanatory Variable
Time based growth
Day of week
Public holiday
0.02
(0.29)
0.10
(2.20)
0.06
(1.39)
Cool day
Mild day
Warm day
TEMP
-0.0035
(-.67)
TEMP2
RAIN
-0.012
(-9.36)
RAIN2
-0.03
(-9.75)
0.0002
(6.11)
-0.08
(-1.84)
-0.41
(-9.83)
Light rain
Heavy rain
-0.09
(-1.94)
-0.40
(-9.84)
-1.21
(-11.28)
Presence of rain_5am-8am
R2
.56
.61
.58
.58
.61
KEY: The associated t stats are presented in parenthesis. The critical values for t stats are for a 95%
confidence interval. Variables which are significant at a 95%confidence level are lightly shaded.
TRB 2012 Annual Meeting
Paper revised from original submittal.
Ahmed, Rose, Figliozzi & Jacob
17
Although ridership varies relatively little across weekdays, Tuesday appears to
have the highest volume. Ridership decreases as the week progresses. On Public holidays,
the daily bicyclist volume decreases by around 70%.
Model 1 which includes only a linear, continuous temperature and rainfall effect
produces the lowest R2 value. Temperature is insignificant in all except model number 3
where only the effect of mild days (18.2 0C to 21.60C) is found to be statistically
significant. Model number 2 produces an optimum riding temperature of 20 0C which falls
into the mild temperature category for Brisbane. While not statistically significant this
result is constant with model number 3 which shows that on days with mild temperature
the ridership is around 10% higher than that on very warm days (temperature >24.30C).
When the effect of light rain is significant (in model number 4) it produces a 9%
decrease in ridership. Heavy rain has an effect four times greater than that of light rain.
When rain enters as a continuous variable (model 1 and 2), each 1mm of increase in
rainfall reduces the bicyclist volume by 1 to 3%.
The positive co-efficient on the variable RAIN2 which was found in case of both
Portland and Brisbane, means the relationship between rainfall and ridership is ‘U’ shaped.
It implies that ridership would initially decrease with increasing rainfall before reaching a
minimum and then increase as rainfall gets heavier. The priori expectation would be that
ridership would continue to decline with increasing rainfall. So, the positive coefficient on
RAIN2 is counter to expectations. To explore the rainfall effect further it was coded
categorically into Light and Heavy rain (models 3 and 4) which as discussed above,
demonstrate the greatest impact of heavy rain.
Previous research conducted in Montreal identified that the bicyclist volume in an
hour is not only affected by the presence of rain in that hour but it also affected if there was
rain in previous hours (1). To examine how presence of rainfall during the morning peak
period (5am-8am) affects daily bicyclist volume in Brisbane, the variable, ‘Presence of
rain_5am-8am’, is included in model 5. Here, rain fall is categorized as ‘No rain’ (total
rain fall during 5am to 8am = 0mm) and ‘Presence of rain_5am-8am’ (minimum total
rainfall during 5am to 8am > 0 mm). The variable is coded as a dummy variable with the
base case corresponding to ‘No rain’. A high statistically significant effect is found which
indicates that if rainfall occurs within the specified period it reduces the daily bicyclist
volume by 70%, compared to when there is no rain. That model also produced the highest
R2 values and thus explained the greatest variation in the ridership data. As the hourly data
were not available for Portland, it was not possible to report results for a comparable
model for Portland.
COMPARISON BETWEEN PORTLAND AND BRISBANE MODEL RESULTS
The models for both cities explain a high proportion of the variation in bicyclist volume.
Across the different model specifications, most of the explanatory variables are found to be
statistically significant. The comparisons discussed below between these two cities are
based on the results from model 3 as most of the explanatory variables for model 3 are
statistically significant across the two locations and the identified effects of weather
variables reflect each location’s base climatic conditions.
Day of the Week Effect
The pattern of ridership across working days is similar in Portland and Brisbane where
Friday is associated with lowest volume in both cases. However, it is noticeable that Friday
gets much lower volume compared to other weekdays in Brisbane (coefficients of the days
in Brisbane models are larger than that for Portland). The results indicate that commuters
TRB 2012 Annual Meeting
Paper revised from original submittal.
Ahmed, Rose, Figliozzi & Jacob
18
in Brisbane are less likely to ride to work in Friday compared to the commuters in
Portland.
Effect of Public Holiday
Public holidays influence commuter bicyclists nearly in the same manner across the cities
though the effect is slightly higher in Brisbane. When it is a public holiday daily bicyclist
volume decreases by 72% in Brisbane compared to 63% in Portland.
Weather Effect
There is evidence of a difference in the sensitivity with respect to temperature across the
two cities. Ridership in Portland reaches its peak on very warm days. On days with other
temperature ranges (cool day, mild day and warm day), estimated ridership is much lower
compared to that on very warm days. The optimum riding temperature in Portland is
24 0C, which is in the ‘very warm’ category for its local climate. On mild days, daily
ridership is 30% lower than that on very warm days. In Brisbane the optimum riding
temperature is 20 0C, which is in the ‘mild’ category for its local condition. Bicyclist
volume is 10% higher on mild days than on very warm days in Brisbane. Thus, very warm
days in Portland stimulate ridership. However, in Brisbane, ridership is much lower on
very warm days compared to other days.
The effect of light rain was not found to be statistically significant for either city.
However, heavy rain is found have significant effect in both cities, though the sensitivity in
the two cities is different. Heavy rain produces a 23% decrease in ridership in Portland
whereas it is 33% in Brisbane.
CONCLUSIONS AND RESEARCH DIRECTIONS
This paper has investigated weather impacts on bicycle ridership in Portland and Brisbane
using a regression modeling approach. Temperature and rainfall, which were previously
identified as two important weather parameters, were again found to significantly influence
ridership.
The models are able to explain a high proportion of variability in daily bicyclist
volume. The study confirms the sensitivity of ridership to weather with different extents
across the cities. Temperature affects ridership both in Portland and Brisbane but with
different degrees which correspond to their base climatic conditions. Moreover, a non liner
effect of temperature is captured in Portland where 240C is appeared to be ideal for riding.
Heavy rain has the greatest impact on ridership in both cities. Furthermore, presence of
rainfall during morning peak period influences daily ridership in Brisbane.
In the future, hourly variations in ridership as a result of changes in weather will be
investigated. Other variables which have the potentiality to influence cyclist volume but
were not incorporated in this analysis will be considered in future work such as effect of
humidity. More sites from both Portland and Brisbane will be analyzed as an extension of
this study. Moreover, disaggregate travel data will be analyzed to examine how bicycle
riders adapt their day to day travel behavior in response to changes in weather. Exploring
disaggregate data will facilitate an understanding of the relative effects of weather on male
versus female riders and the extent to which riders who have invested in appropriate
equipment (mud guards, weather proof panniers) are less sensitive to changes in weather.
From a transport policy perspective, it would be useful for future research to provide
insight into whether riders who have access to end of trip facilities (change rooms,
showers, lockers, airing closets etc) are less sensitive to changes in weather. Furthermore,
the modal shift of bicycle riders as a result of weather changes will be quantified which
TRB 2012 Annual Meeting
Paper revised from original submittal.
Ahmed, Rose, Figliozzi & Jacob
19
will provide insight into whether increases in public transportation ridership or additional
pressure on the road system through increased use of private motorized transport might be
expected when weather is not favorable for riding.
ACKNOWLEDGEMENT
The authors wish to thank Peter Berkeley, Mark Dorney and Michael Langdon from
Department of Transport and Main Roads, Queensland and City of Portland who provided
invaluable data and support. Any opinions, findings and conclusions or views expressed
here are those of the authors.
REFERENCES
1. Garrard, J., S. Crawford, and N. Hakman. Revolutions for Women: Increasing Women`s
Participation in Cycling for Recreation and Transport. Deakin University, School of
Health and Social Development, 2006, p. 8.
2. Dill, J. Bicycling for Transportation and Health: The Role of Infrastructure. Journal of
Public Health Policy, Vol. 30, 2009, pp. 95-110.
3. Moreno, L., and T. Nosal. Weather or Not to Cycle; Whether or Not Cyclist Ridership
Has Grown: A Look at Weather’s Impact on Cycling Facilities and Temporal Trends in
an Urban Environment.CD-ROM. Transportation Research Board of the National
Academies, Washington, D.C., 2011.
4. Seneinejad, S., C. Kennedy, and M. J. Roorda. Modelling the Impact of Weather on
Active Transportation. 12th World Conference on Transport Research, Lisbon, Portugal,
July 11 to 15, 2010.
5. Climate Change in Australia, technical report, 2007. CSIRO and Bureau of
Meteorology, Australia.
6. Dill, J., and K. Voros. Factors Affecting Bicycling Demand. In Transportation
Research Record: Journal of the Transportation Research Board, No. 2031,
Transportation Research Board of the National Academies, Washington, D.C., 2007, pp.
9–17.
7. Ride To Work Day, Event Evaluation Report. Bicycle Victoria, Australia, 2009.
8. Rose, G., and H. Marfurt. Travel Behavior Change Impacts of a Major Ride to Work
Day Event, Transportation Research Part A, Vol. 41, 2007, pp. 351–364.
9. Ahmed, F., G. Rose, and C. Jacob. Impact of weather on commuter cyclist behavior and
implications for climate change adaptation.CD-ROM. 30th Australasian Transport
Research Forum, Canberra: Forum Papers. Canberra, Australia, September 29 to
October 1st.
10. Maus, J. Portlands Earns Platinum; Becomes First Major U.S City to Win the Award.
Bike Portland, April 29, 2008.
www.bikeportland.org/2008/04/29/portland-gets-platinum-becomes-first-major-uscity-to-win-the-award/ Accessed July 28, 2010.
11. League of American Bicyclists, New York, Washington, D.C.
www.bikeleague.org/programs/bicyclefriendlyamerica/communities/index.php.
Accessed July 28, 2010.
12. Queensland cycle strategy implementation report, 2008-2009.Department of Transport
and Main Roads, Queensland, Australia.
13. Australian bicycle council. Annual report 2009. Australia.
14. Michael, L. Personal Communication. Department of Transport and Main Roads,
Queensland, Australia, 2011.
15. Bureau of Meteorology, Australian Government.
TRB 2012 Annual Meeting
Paper revised from original submittal.
Ahmed, Rose, Figliozzi & Jacob
20
www.bom.gov.au/global. Accessed July 12, 2011.
16. Nankervis, M. The Effect of Weather and Climate on Bicycle Commuting,
Transportation Research Part A, Vol. 33, 1999, pp. 417-431.
17. Thomas, T., R. Jaarsma, and B. Tutert. Temporal Variations of Bicycle Demand in the
Netherlands: Influence of Weather on Cycling. Transportation Research Board 88th
Annual Meeting. CD-ROM. Transportation Research Board of the National
Academies, Washington, D.C., 2009.
18. Lewin, A. Temporal and Weather Impacts on Bicycle Volumes.CD-ROM.
Transportation Research Board of the National Academies, Washington, D.C., 2011.
19. Rose, G., F. Ahmed, M. Figliozzi, and C. Jakob. Quantifying and Comparing the
Effects of Weather on Bicycle Demand in Melbourne (Australia) and Portland (USA),
CD-ROM. Transportation Research Board of the National Academies, Washington,
D.C., 2011.
20. Richardson, A. J. Seasonal and Weather Impacts on Urban Cycling Trips. TUTI Report
1-2000, The Urban Transport Institute, Victoria, 2000.
21. Phung, J., and G. Rose. Temporal Variations in Usage of Melbourne’s Bike Paths.
Proceedings of 30th Australasian Transport Research Forum, Melbourne: Forum
Papers. CD-ROM. Melbourne, Victoria, Australia, 25-27 September 2007, p. 1CD
ROM.
22. Khattak, A., and A. Palma. The Impact of Adverse Weather Conditions on the
Propensity to Change Travel Decisions: A Survey of Brussels Commuters.
Transportation Research Part A, Vol. 31, No. 3, 1997, pp. 181–203.
23. Aultman-Hall, L., D. Lane, and R. R. Lambert. Assessing Impact of Weather and
Season on Pedestrian Traffic Volumes. In Transportation Research Record: Journal of
the Transportation Research Board, No. 2140, Transportation Research Board of the
National Academies, Washington, D.C., 2009, pp. 35-43.
24. Rose, G. Transit Passenger Response: Short and Long Term Elasticities Using Time
Series Analysis, Transportation, Vol. 13, 1986, pp. 131-141.
25. Bicycles Sensors, Eco Counter.
www.eco-compteur.com/Bicycles-Sensors.html?wpid=15037. Accessed July 27, 2010.
26. Ahmed, F., G. Rose, and C. Jacob. Sensitivity of Commuter Cyclists to Changes in
Weather in Queensland. Asia-Pacific Cycle Congress, Brisbane, 2011.
TRB 2012 Annual Meeting
Paper revised from original submittal.